Abstract
Geographic atrophy (GA), an advanced form of age-related macular degeneration, is characterized by progressive loss of photoreceptors (PR) and subsequent degeneration of the retinal pigment epithelium (RPE). Accurate identification and quantification of GA are essential for patient stratification and monitoring. Current clinical trial primarily relies on manual delineation of GA lesions from two-dimensional fundus autofluorescence images, a time-consuming process that may miss subtle three-dimensional morphologic changes in retinal architecture. We present GAIZ, a morphology-aware deep learning pipeline incorporating a hybrid loss function that includes a novel retinal-layer ordering loss and two biomarker-specific losses (PR and RPE disruption losses) to detect and quantify GA biomarkers. In addition, it leverages volumetric spectral-domain optical coherence tomography (SD-OCT) to better capture micro-structural degeneration patterns and the pipeline comprises of a vision state-space model that integrates hierarchical feature representation with the efficient long-range modeling capabilities of Mamba state-space layers. We evaluated GAIZ on a multi-center dataset containing expert annotations of retinal layers, from which we derive the aforementioned biomarkers. Our method achieves an ICC of 0.98 [0.97, 0.99] and 0.92 [0.86, 0.95] for RPE and PR loss area measurements, respectively. Additionally, GAIZ also demonstrates strong performance in secondary metrics, including mean intersection-over-union and Dice score for inter-retinal layer segmentation. These results show that GAIZ effectively captures the spatial progression of PR and RPE loss, providing a strong foundation for automated OCT-based assessment and future longitudinal modeling of macular disease progression.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_029.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=6hlHfFsxr3
BibTex
@InProceedings{KamSha_GAIZ_MICCAISAT2026,
author = { Kamran, Sharif Amit AND Cheung, Matt Y. AND Parmar, Chaitanya AND Luu, Jennifer Y. AND Gale, Sara AND Di, Junrui AND Standish, Kristopher A. AND Govind, Darshana},
title = { { GAIZ: Automated Detection of Geographic Atrophy Biomarkers via Volumetric OCT using Novel Morphological Objectives } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17272},
month = {pending},
page = {pending}
}